Re: Selectively suppress -Wcast-function-type (issue 357770043 by address@hidden)

[dak]

On 2018/08/13 12:29:46, Dan Eble wrote:
> On 2018/08/13 07:11:38, dak wrote:
> > I'd use scm_func_cast<scm_t_blabla> (actual_func)

> I'll do that.

> > Any chance that an implementation via a double cast (with something
like void
> *
> > in between) will work?  Strictly speaking that's worse than casting
from one
> > function pointer to another since function and data pointers need
not be the
> > same size on segmented architectures, but lots of code does stuff
like that
> > anyway.

> The first thing I tried was an intermediate (void*), and that did
eliminate the
> warning.  The reason you cite is the reason I didn't propose it, but
if you're
> comfortable with it, I'll do it.  Interaction with Guile is not a
place where I
> care to assert my preferences.

Personally, I think that we should rather disable this warning globally.
 An explicit cast is an explicit cast.  There would be a weak argument
that a C-style cast may expect an actual working conversion to take
place (and I have little enough sympathy with that argument), but an
explicit reinterpret_cast quite clearly states "shut up, I know what I
am doing".  So I don't really think that it's warranted to tap-dance
around GCC here and I'd rather stop GCC on the command line from
flagging this.

Apropos: what have you done about the warnings regarding SCM_UNPACK ?
Those use something like (0 ? *(SCM*)0 : ...) for type coercion/checking
and the resulting warnings clutter everything for me.  I'll likely
change my local Guile-1.8 source for quieting those.

https://codereview.appspot.com/357770043/